tune.par.vsflcm: tune key parameters(K,lambda,lam.nuc,lam.smo) using cross...

In Hongming-Pu/vsFlcm: Variable Selection(Group Lasso) for the Functional Linear Concurrent Model

Description

tune key parameters(K,lambda,lam.nuc,lam.smo) using cross validation in vsflcm, if you don't want to tune the parameter by yourself, this is for you.

Usage

tune.par.vsflcm(formula, data = NULL, id.time = NULL, id.sub = NULL,
  t.min = NULL, t.max = NULL, seed.split = 4, num.fold = 3,
  thre = 0.9, fpc.on = TRUE, intercept = TRUE, K = NULL,
  lambdas = 10^seq(-2, 2, 1), lam.smos = c(0, 10^seq(-2, 2, 1)),
  lam.nucs = c(0, 10^seq(-2, 2, 1)), K0 = 2, spline.fun = "B-spline",
  method.obj = "nuclear", K.max = 10, delta = 0.1,
  method.optim = "BFGS", times = 1, spline.fun2d = NULL, maxit = 200)

Arguments

formula	formula for the regression. should have form y ~ V1 + V2 + ... + Vk. Note: don't contain id.time
data	data frame
id.time	the variable that represents time
id.sub	variable giving subject ID
t.min	minimum value to be evaluated on the time domain. if 'NULL', taken to be minium observed value.
t.max	maximum value to be evaluated on the time domain. if 'NULL', taken to be minium observed value.
seed.split	seed to randomly split the data set into training set and test set
num.fold	num.fold-fold cross validation
thre	when automaticly chosing K, K will be chosen as the smallest one which ensures that the model can explain thre of the total variance.
fpc.on	Logical. Whether fpc will be used
intercept	logical, whether intercept is used
K	number of spline basis functions for coefficients and fpc. If default this function will automaticly choose K. Otherwise you need to provide a value.
lambdas	tuning parameter vector for lambda to choose from
lam.smos	tuning parameter vector for lam.smo to choose from
lam.nucs	tuning parameter vector for lam.nuc to choose from
K0	number of FPCs
spline.fun	spline basis functions. If not default, it should be a function taking a vector(corresponding to time) belonging to [0,1] as input and returning a length(vector)*K spline matrix where each row corresponds to a time point in the vector and each column corresponds to a spline basis function
method.obj	the methodology chosen. Should be 'nuclear' or 'nonconvex'. Typically, we recommend 'nuclear'
K.max	max K can be chosen automaticly, only useful when K is NULL
delta	Only useful when method.obj='nonconvex'. When the square of the norm of the coefficient is smaller than delta, L1 penalty will smoothly turned into L2 penalty
method.optim	the optimization method to be used. For details and other options, please refer to optim
times	Only useful when method.obj='nonconvex'. the algorithm randomly chooses times initial points to optimize and select the best
spline.fun2d	second derivative of spline.fun, if spline.fun is default then the second derivative will be computed automaticly and you don't need to provide values for spline.fun2d. Otherwise you need to provide a function with input and output dimensions in accordance with spline.fun if you want to model the smoothness.
maxit	maxit of optim

Author(s)

Hongming Pu [e-mail]

Hongming-Pu/vsFlcm documentation built on May 28, 2019, 12:41 p.m.